NAME¶

netperfmeter — Network Performance Meter

SYNOPSIS¶

DESCRIPTION¶

netperfmeter is a network performance meter for the TCP, MPTCP, UDP, DCCP and SCTP transport protocols over IPv4 and IPv6. It simultaneously transmits bidirectional flows to an endpoint and measures the resulting flow bandwidths and QoS. The results can be written as scalar files (summary of the run) and vector files (details per frame). These files can be processed further, e.g. for analysis and plotting of the results.

ARGUMENTS¶

The following arguments can to be provided:

local_port

The port number for the passive side's data socket for TCP, UDP, DCCP, SCTP. For MPTCP, the port number will be local_port-1, to allow for both, TCP and MPTCP connections. The port number of the control socket will be local_port+1. Specifying a port number turns NetPerfMeter into passive mode, i.e. it will wait for incoming connections.

remote_endpoint:remote_port

Specifies the passive endpoint to connect to (number as given for local_port on the passive instance). Specifying remote_endpoint:remote_port turns NetPerfMeter into active mode, i.e. it will connect to the specified remote endpoint.

x | --control-over-sctp

On the passive side, this option enables accepting control connections via SCTP (default). It enables SCTP in addition to other protocols for the control transport. On the active side, this option uses SCTP for the control connection to the passive side. This is the default, but may cause connection problems in restricted networks or due to NAT traversal.

-X | --no-control-over-sctp

On the passive side, this option disables incoming control connections via SCTP.

-y | --control-over-tcp | -Y | --no-control-over-tcp

On the passive side, this option enables accepting control connections via TCP (default). It enables TCP in addition to other protocols for the control transport. Note: By default, both -- SCTP and TCP -- are enabled on the passive side. On the active side, this option uses TCP for the control connection to the passive side. This may be necessary for NAT traversal or due to restrictive firewalls.

-Y | --no-control-over-tcp

On the passive side, this option disables incoming control connections via TCP. This implicitly also disables control connections over MPTCP.

-w | --control-over-mptcp

On the passive side, this option enables accepting control connections via MPTCP. This implicitly also enables control connections over TCP! On the active side, this option uses MPTCP for the control connection to the passive side.

-W | --no-control-over-tcp

On the passive side, this option disables incoming control connections via MPTCP. However, it does not deactivate TCP!

-L address[,address,...] | --local address[,address,...]

Specifies address(es) of the local *data* endpoint. For TCP, UDP and DCCP, only the first address is used! On an active node, this option applies to the following flows, i.e. it must be set before specifying a flow!

-l address[,address,...] | --controllocal address[,address,...]

Specifies address(es) of the local *control* endpoint (SCTP or TCP). For TCP, only the first address is used!

-K key_file | --tls-key key_file

Specifies the TLS key file for QUIC authentication. This option is used by the passive side.

-J certificate_file | --tls-cert certificate_file

Specifies the TLS certificate file for QUIC authentication. This option is used by the passive side. The certificate file must correspond to the key file provided by --tls-key. The certificate must be for server usage (X.509 Extended Key Usage: TLS Web Server Authentication).

-I ca_certificate_file | --tls-ca ca_certificate_file

Specifies the TLS root CA certificate file for QUIC authentication.

-H hostname_file | --tls-hostname hostname

Specifies the TLS certificate hostname for QUIC authentication. It is used by the active side. The hostname must match the remote (i.e. passive side's) hostname, as provided by the passide side's TLS certificate, to allow for hostname verification. As alternative, it is possible to add the IP addresses into the X.509 SubjectAltName of its certificate.

--6 | --v6only

Use sockets with IPv6 only (by setting IPV6_V6ONLY socket option).

--display

Display live I/O statistics, updated every second. This is enabled by default.

--nodisplay

No not display live I/O statistics. This is useful for running NetPerfMeter in background as a service.

-o bytes | --sndbuf bytes

Sets the sender buffer size to the given number of bytes.

-i bytes | --rcvbuf bytes

Sets the receiver buffer size to the given number of bytes.

-N count | --count count

Add all specified flows the given number of times. That is, given a count of 5, each specified flow will be added 5 times. This allows one to easily create multiple flows with the same parameters.

-T seconds | --runtime seconds

Specifies the measurement runtime in seconds. After the given time span, NetPerfMeter will finish the measurement.

-C configuration_file | --Ar -config -configuration_file

Specifies the name of the configuration file to write. The configuration file will contain the used flow parameters.

-S scalar_file_pattern | --scalar scalar_file_pattern

Specifies the name pattern of the scalar files to write. If the suffix of this name is .bz2, the file will be BZip2-compressed on the fly. The scalar name is automatically extended to name the flow scalar files by adding -<active|passive>-<flow_id>-<stream_id> before the suffix. For example for scalar.vec.bz2, the name of the scalar file for flow 5, stream 2 on the active node will be scalar-active-00000005-0002.vec.bz2.

-V vector_file_pattern | --vector vector_file_pattern

Specifies the name pattern of the vector files to write. If the suffix of this name is .bz2, the file will be BZip2-compressed on the fly. The vector name is automatically extended to name the flow vector files by adding -<active|passive>-<flow_id>-<stream_id> before the suffix. For example for vector.vec.bz2, the name of the vector file for flow 5, stream 2 on the passive node will be vector-passive-00000005-0002.vec.bz2.

-A description | --activenodename description

Sets a textual description of the active node (e.g. Client).

-P description | --passivenodename description

Sets a textual description of the passive node (e.g. Server).

--loglevel level

Sets the minimum logging level to the provided value: 0=trace, 1=debug, 2=info, 3=warning, 4=error, 5=fatal. Default: 2 (info).

--logcolor on|off

Enables (on) or disables (off) the usage of ANSI color escape sequences to colorise the log output according to its level. Default: on.

--logappend file

Sets a file name for appending the log output to. If a log file is set, no log output is made to the console (stderr). Otherwise, all log output will be written to the console.

--logfile file

Sets a file name for writing the log output to. Existing content is truncated. If a log file is set, no log output is made to the console (stderr). Otherwise, all log output will be written to the console.

-q | --quiet

Sets the logging level to 4 (error).

-! | --verbose

Sets the logging level to 0 (trace).

-t FLOWSPEC | --tcp FLOWSPEC

Establish a new TCP connection. The flow of this connection must be specified by a FLOWSPEC specification as following parameter. For backwards compatibility, the option cmt=mptcp creates an MPTCP connection instead of a TCP connection.

-m FLOWSPEC | --mptcp FLOWSPEC

Establish a new MPTCP connection. The flow of this connection must be specified by a FLOWSPEC specification as following parameter. Note: DCCP is not available on all platforms, yet. Currently, only Linux provides MPTCP in its official kernel.

-u FLOWSPEC | --udp FLOWSPEC

Establish a new UDP connection. The flow of this connection must be specified by a FLOWSPEC specification as following parameter.

-d FLOWSPEC | --dccp FLOWSPEC

Establish a new DCCP connection. The flow of this connection must be specified by a FLOWSPEC specification as following parameter. Note: DCCP is not available on all platforms, yet. Currently, only Linux provides DCCP in its official kernel.

-s FLOWSPEC | --sctp FLOWSPEC [FLOWSPEC ...]

Establish a new SCTP association. The streams of this association must be specified by one or more FLOWSPEC specifications as following parameters.

-k FLOWSPEC | --quic FLOWSPEC [FLOWSPEC ...]

Establish a new QUIC connection. The streams of this connection must be specified by one or more FLOWSPEC specifications as following parameters.

FLOWSPEC

Specifies a new flow. The format is: outgoing_frame_rate:outgoing_frame_size:incoming_frame_rate:incoming_frame_size:option:... The first four parameters (outgoing_frame_rate:outgoing_frame_size:incoming_frame_rate:incoming_frame_size) may be substituted by the option "default", creating a flow with some more or less useful default parameters.

outgoing_frame_rate

The frame rate of the outgoing transfer (i.e. active node to passive node). If set to const0, the sender will be saturarted, i.e. it will try to send as much as possible.

outgoing_frame_size

The frame size of the outgoing transfer (i.e. active node to passive node). If set to const0, there will be *no* data transmission in this direction.

incoming_frame_rate

The frame rate of the incoming transfer (i.e. passive node to active node). See outgoing_frame_rate for details.

incoming_frame_size

The frame size of the incoming transfer (i.e. passive node to active node). See outgoing_frame_size for details.

Random distribution:

Frame rate and frame size may be distributed randomly, using different random distributions. A new random value is drawn for the next frame send time or next frame size. The following distributions are supported:

const<value>

Constant, i.e. always the same. Example: a frame rate const1000 means that all frames have a size of 1000 bytes.

uniform<a>,<b>

Uniform distribution from the interval [a,b). Example: uniform900,1100.

exp<mean>

Exponential distribution with mean <mean>. Example: exp1000.

pareto<location>,<shape>

Pareto distribution with location <location> and shape <shape>. Example: pareto0.166667,1.5.

normal<mean>,<stddev>

Normal distribution with mean <mean> and standard deviation <stddev>. Example: normal1000,500. Note that normal distribution may result in negative values, which may be ignored (frame size) or interpreted as saturated traffic (frame rate). Use truncnormal to prevent negative values.

truncnormal<mean>,<stddev>

Truncated normal distribution with mean <mean> and standard deviation <stddev>, i.e. normal distribution without negative values. Example: truncnormal1000,500.

Possible flow options:

id=Flow Identifier

Sets an ID number for the flow. The IDs within a measurement must be unique!

description=Description

Sets a textual description of the flow (e.g. HTTP-Flow). Do not use spaces in the description!

maxmsgsize=Bytes

Splits frames into messages of at most the given number of bytes. Messages may not exceed 65535 bytes.

defragtimeout=Milliseconds

Messages not received within this timeout after the last successfully received message are accounted as lost. NOTE: this also happens if the transport protocol is reliable and the message is actually received later!

unordered=Fraction

Specifies the fraction of the messages that will be sent in unordered mode (SCTP only).

ordered=Fraction

Specifies the fraction of the messages that will be sent in ordered mode (SCTP only).

reliable=Fraction

Specifies the fraction of the messages that will be sent in reliable mode (SCTP only).

unreliable=Fraction

Specifies the fraction of the messages that will be sent in unreliable mode (SCTP only).

rtx_timeout=Milliseconds

Sets the retransmission timeout for unreliable messages (SCTP only).

rtx_trials=Trials

Sets the retransmission trials for unreliable messages (SCTP only).

onoff=t1,t2,...[,repeat]

A list of time stamps when the flow should be activated or deactivated. If onoff is given, the flow is off at startup. At t1, it will be turned on; at t2, it will be turned off, etc.. Time stamps can be given as absolute values (e.g. onoff=0,10,30 - to turn on at t=0, turn off at t=10 and turn on again at t=30 until end of measurement) or relative values (e.g. on=10,+30,+60 - to turn on at t=10, turn off at t=40 and turn on again at t=100 until end of measurement). A repetition of the list is possible with the keyword "repeat" at the end of the list. Then, all values need to be relative values and the number of items must be even.

error_on_abort=on|off

By default, the active side stops with an error when a transmission tails (e.g. on connection abort). This parameter turns this behaviour on or off.

nodelay=on|off

Deactivate Nagle algorithm (TCP and SCTP only; default: off).

debug=on|off

Set debug mode for flow (default: off). Note: this is for debugging and testing NetPerfMeter only; it (usually) has no function!

v6only

Use socket with IPv6 only (by setting IPV6_V6ONLY socket option).

cc=name

Set congestion control name (TCP and MPTCP for Linux only).

cmt=off|cmt|cmtrpv1|cmtrpv2|like-mptcp|mptcp-like|mptcp

Configures usage of Concurrent Multipath Transfer (CMT): off (turned off; default), cmt (independent paths), cmtrpv1 (CMT/RPv1), cmtrpv1 (CMT/RPv2), mptcp/like-mptcp/mptcp-like (MPTCP), 0-255 (custom value). Currently only supported by CMT-SCTP on FreeBSD systems and MPTCP on Linux systems. Note: CMT for MPTCP always uses MPTCP congestion control.

-h | --help

Prints command help.

-v | --version

Prints program version.

EXIT STATUS¶

The netperfmeter tool exits with 0 on success, and >0 in case of an error.

EXAMPLES¶

Some netperfmeter usage examples:

netperfmeter 9000

Start in passive mode, i.e. accepting connections, on port 9000.

netperfmeter 172.31.33.2:9000 -config=output.config -vector=output.vec.bz2 -scalar=output.sca.bz2 -sctp const5:exp1000:const3:exp500:description="Alpha":onoff=+10 const5:exp1000:const3:exp500:description="Beta":onoff=+30 -tcp const5:exp1000:const3:exp500:description="Gamma":onoff=+60 -runtime=300

Start in active mode, i.e. establish connection to 172.31.33.2, port 9000. Frames will be segmented into messages of up to 4096 bytes. Write configuration to output-*.config. Write vectors to output-*.vec.bz2 (BZip2-compressed). Write scalars to output-*.sca.bz2 (BZip2-compressed). Establish SCTP association. At t=10s, start flow "Alpha" as first stream of this association, using a frame rate of 5 (constant) and average frame size of 1000 bytes (negative exponential distribution) for the outgoing data and a frame rate of 3 (constant) and average frame size of 500 bytes (negative exponential distribution) for the incoming data. At t=30s, start flow "Beta" as second stream on the SCTP association (same parameters as flow "Alpha"). Start TCP connection. At t=60s, start flow "Gamma" over the TCP connection (same parameters as flow "Alpha"). At t=300s, stop the measurement.

netperfmeter 172.31.33.2:9000 -control-over-tcp -tcp const2:const1000

Start in active mode, i.e. establish connection to 172.31.33.2, port 9000. The control connection uses TCP instead of SCTP.

netperfmeter 172.31.33.2:9000 -sctp const0:const0:uniform20,30:uniform900,1100

SCTP traffic with some variation.

netperfmeter 172.31.33.2:9000 -udp const0:const0:const25:const2500:onoff=+pareto0.166667,1.5,+pareto0.166667,1.5,repeat

Some on-off background traffic with Pareto distribution.

netperfmeter 9000 -tls-key my-server.key -tls-cert my-server.crt -tls-ca my-ca.crt

Start in passive mode, with TLS setup for QUIC. See https://www.nntb.no/~dreibh/netperfmeter/#quic-communication for details about QUIC usage!

netperfmeter 172.31.33.2:9000 -tls-hostname my-server.domain -tls-ca my-ca.crt -quic const10:const128:const25:const1200

Start in active mode, i.e. establish connection to 172.31.33.2, port 9000. Set TLS server name of 172.31.33.2 for QUIC to my-server.domain. See https://www.nntb.no/~dreibh/netperfmeter/#quic-communication for details about QUIC usage!

netperfmeter -version

Shows the version.

netperfmeter -help

Shows usage help.

SEE ALSO¶

combinesummaries(1), createsummary(1), netperfmeter(1), netperfmeter-module-loader(1), plot-netperfmeter-results(1), runtimeestimator(1)

NOTES¶

This program is part of NetPerfMeter. The latest version of NetPerfMeter can be found on the NetPerfMeter Homepage at NetPerfMeter Homepage.

AUTHORS¶

Thomas Dreibholz, Homepage